<result> <column_name>deep_sleep_pct</column_name> <family>uncertain</family> <subtype>none</subtype> <confidence>0.5</confidence> <requires_user_confirmation>true</requires_user_confirmation> <sum_meaningful>false</sum_meaningful> <signed>false</signed> <reducers> <primary_reducer>mean</primary_reducer> <valid_reducers>mean, min, max, median, distribution</valid_reducers> <excluded_reducers>sum (a percentage share of sleep does not accumulate), product (no sequence to compound along), geomean (values are not ratios clustered near 1.0)</excluded_reducers> <average_semantics>arithmetic_mean</average_semantics> </reducers> <inner_outer_measure> <inner_measure_reducer>min</inner_measure_reducer> <outer_measure_reducer>max</outer_measure_reducer> </inner_outer_measure> <multiplicative_partner_derivable>false</multiplicative_partner_derivable> <reasoning> The table is a DIMENSION with no sequence axis, so multiplicative is highly implausible and no cumsum or cumprod reconstruction is possible. The "pct" token is the only multiplicative signal. The data contradict it: mean and median are about 21.8, nowhere near 1.0, all values are strictly positive within 8.1 to 28.0, and log_symmetry_gain is negative. Because naming and numeric evidence conflict and there is no sequence evidence, I abstain on family. The column behaves like a per-entity bounded intensity (a share of sleep time), so mean is the safest rollup. Sum is not meaningful. I flag it for human review to confirm intensive_additive. </reasoning> <table_type_conflict>false</table_type_conflict> </result>